Top Banner
A wireless sensor network for traffic surveillance University of California, Berkeley Sing Yiu Cheung, Sinem Coleri, Ram Rajagopal, Pravin Varaiya
51

A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

Jun 13, 2018

Download

Documents

dinhdiep
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Page 1: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

A wireless sensor network for

traffic surveillance

University of California, Berkeley

Sing Yiu Cheung, Sinem Coleri, Ram Rajagopal,

Pravin Varaiya

Page 2: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

2

Outline

• Traffic measurement

• Wireless Sensor Networks

• Vehicle Count, Occupancy, Speed

• Vehicle Classification

• Vehicle Re-identification

• Communication protocol

• Future work

Page 3: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

3

Traffic Applications

Stop-BarDetection

170

Traffic SignalControl

Freeway RampMetering

Advance Detection

ParkingGuidance

ParkingEnforcement

Freeway TrafficData

170

SystemCounts

Page 4: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

4

Freeway Traffic Management

ACCIDENT!

Measurement Backhaul Processing

Internet

Control & Info

Cellular

Traffic Management

Center

TrafficControl

PeMShttp://pems.eecs.berkeley.edu

Page 5: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

5

Intersection Signal Control

Source: Mirchandani, P. and L. Head, ‘Rhodes’ Traffic-adaptive control system, TRB 2001

Adaptive Traffic Signal Control System

Measurements: detectors & signals Actuator signals

raw

data

feedback &

decisions

Page 6: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

6

Data Requirements

• Freeway detectors must report count,

occupancy, speed every 30 s, 95%

accuracy

• Intersection detectors must report vehicle

presence at intersection within 0.1 s, 99%

accuracy

• Parking detectors must report events

within one minute, 90-99% accuracy

Page 7: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

7

Current Traffic Measurement

Technologies

Inductive loop Video Microwave radar

• Loop detector is the standard; lasts years but unreliable;

closing lane to cut loops in pavement is disruptive;

alternatives are microwave radar, video

• Can wireless sensor networks compete?

Page 8: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

8

Outline

• Traffic measurement

• Wireless Sensor Networks

• Vehicle Count, Occupancy, Speed

• Vehicle Classification

• Vehicle Re-identification

• Communication protocol

• Future Work

Page 9: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

9

Wireless Magnetic Sensor Networks

•Sensor nodes

•Detect vehicles by change in earth’s magnetic field

•Transmit data to access point via radio

•Access point reports to signal controller or TMC

•Low cost, non-intrusive, flexible, easy to install

Sensor node

Access point

IntersectionFreeway

Page 10: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

10

Prototype sensor Node

Protective cover

can withstand

9000kg

3-axis magnetometer,

microprocessor,

radio transceiver,

antenna, battery

Page 11: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

11

Sensys Vehicle Detector Station

Note: Varaiya is a director of Sensys Networks, Inc

Hardened plastic

mounted flush with

pavement surface,

10 year battery life

GPS receiver,

GPRS/CDMA

interface, poE, or

power over RS485

2”

Page 12: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

12

Advantages of Wireless Sensing

Page 13: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

13

Advantages of Wireless Sensing

Page 14: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

14

Outline

• Traffic measurement

• Wireless Sensor Networks

• Vehicle Count, Occupancy, Speed

• Vehicle Classification

• Vehicle Re-identification

• Communication protocol

• Future work

Page 15: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

15

Magnetometer: Basic principle

Earth’s magnetic field

Ferrous object

Distorted field

HMC1051Z

Page 16: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

16

Vehicle Detection (count)

• Note sharp, localized threshold crossing

runnin

g a

vera

ge

seconds

z axis measurement as vehicle goes over node

vehicle signal

detection signal

Page 17: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

17

Occupancy and speed by Node Pair

Configuration

vehicle signal

from node 1

detection signal

from node 1

vehicle signal

from node 2

detection signal

from node 2occupancy

speed = 6’/time

Page 18: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

18

CCIT test results on I-80 (1/3)

Page 19: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

19

CCIT test results on I-80 (2/3)

Page 20: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

20

CCIT test results on I-80(3/3)

Page 21: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

21

Outline

• Traffic measurement

• Wireless Sensor Networks

• Vehicle Count, Occupancy, Speed

• Vehicle Classification

• Vehicle Re-identification

• Communication protocol

• Future work

Page 22: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

22

FHWA 13 Classes

• Classify detected vehicle from signal

Page 23: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

23

Classification approach

• Collect labeled samples, (x1,y1), …, (xN,yN),xi Rn(i), class of xi = yi {1, … , 13}

• Train good classifier, f: xi yi

• Classify new sample x as f(x)

Page 24: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

24

Difficulties

• Large and unequal sample vector size:

– Vehicle length 5-20 m, speed 2-25 m/s, so

duration is 0.2-10s, fs = 128/256 Hz, 3-axes, so

size n(i) is 75-7680 samples; inter-vehicle

spacing 2-4 s

• Approach: Convert to fixed vector size n

– Preprocessing: h: xi Rn(i) Rn all i

– Extract features (compress): f: Rn Rk, k << n

• Train classifier: g: Rk {1, … , 13}

• Overall: f = g h: Rn(i) {1, … , 13}

Page 25: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

25

Preprocessing + compression A

1. Normalize sample vector to zero mean,

[-1,1] and fixed sample size n: if n(i) > n

downsample; if n(i) < n upsample

2. Extract features by principal components:If X is M n matrix of M test sample

vectors {xi} with SVD X = USVT, XTX =

VS2V, project xi onto subspace of k

eigenvectors with 90-99% of variation

3. Computationally expensive steps

Page 26: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

26

Example A (n=2971)

# components

% v

ari

atio

n

Page 27: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

27

Preprocessing + compression B

1.Replace sample vector xi of size n(i) by

vector of fixed size k (k=10, 20) with

components = average value of xi in bins of

size n(i)/k

2.Extract principal components if necessary

3.Eliminates speed variation, but also length

Page 28: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

28

Example B (k=10)

Page 29: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

29

Preprocessing + compression C

1. Extract (-1,0,1) ‘hill pattern’

2. Add zero padding to obtain (-1,0,1) vector

of fixed size k

Page 30: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

30

Example C

• Very high compression

Page 31: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

31

Classification Schemes

1. Support vector machines (SVM) using

radial basis function kernel

k(u,v) = exp(- |u-v|2)

2. K-Nearest-Neighbor (KNN) scheme

• Several variants depending on distance and

‘voting’ schemes

Page 32: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

32

KNN results (1/4)

• 839 vehicles

• 50% from each class are randomlyselected for training; 50% left for testing

• Correct rate is average from 10 trials

• Length information is not used

Class 2 5 6 8 9 11

# 85 359 45 83 194 73

Trucks

Page 33: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

33

KNN results (2/4)

• using length may improve accuracy

Page 34: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

34

KNN results (3/4)[%] Classified as Counts

Observed Class 5 6 8 9 11

5 0.84 0.07 0.05 0.03 0.01 217

6 0.60 0.23 0.06 0.11 0.00 23

8 0.42 0.03 0.22 0.29 0.04 53

9 0.18 0.03 0.17 0.55 0.08 118

11 0.07 0.02 0.07 0.36 0.49 20

Avg. Counts 241.6 24.7 44.5 97.6 22.6

FWHA Class 5 6 8 9 11 Total

# 435 47 106 236 40 864

• Average correct rate is 64%

Page 35: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

35

KNN results (4/4)Distribution of Magnetic Length

4

5

6

7

8

9

10

11

12

0 5 10 15 20 25 30 35 40

Estimated Magnetic Length [m]

FW

HA

Cla

ss

FHWA Classes

Magnetic Length [m] 5 6 8 9 11

Mean 10.35 11.22 17.89 20.79 21.43

Std Dev 3.21 3.36 6.31 6.11 5.41

Page 36: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

36

Outline

• Traffic measurement

• Wireless Sensor Networks

• Vehicle Count, Occupancy, Speed

• Vehicle Classification

• Vehicle Re-identification

• Communication protocol

• Future work

Page 37: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

37

Vehicle re-identification

• Use an array of nodes to

compensate for offset

• Experiment conducted with 7

test vehicles– Buick Le Sabre 97 (x 2)

– Toyota Corolla 89

– LandRover Range Rover 96

– Ford Taurus 96

– Ford Taurus 2000

– Ford WindStar (Van)

• Each vehicle driven over SN

array 5 times: Note 2 vehicles

are the same model

Page 38: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

38

Compress with 10 ave bar

• No length or speed

information

• Compare correlation of

the Average-Bar from

all three axes

• 100% correct re-

identification among all

7 test vehicles

Compressed data from same vehicle

Page 39: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

39

Summary

• Wireless magnetic sensor network has

better detection properties than

alternatives

• What about lifetime?

• Depends on energy consumption

Page 40: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

40

Outline

• Traffic measurement

• Wireless Sensor Networks

• Vehicle Count, Occupancy, Speed

• Vehicle Classification

• Vehicle Re-identification

• Communication protocol

• Future work

Page 41: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

41

Energy Calculations

• Focus on radio energy

• Suppose battery rating is 2 Amp-hr @ 3V

• In 1 year = 8700 hours, average power

budget is

– 230 μA for 1 year lifetime

– 23 μA for 10 year lifetime

• What kind of protocol and data rate can be

supported with this power budget?

Page 42: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

42

Mica2 CSMA protocol

• Mica2 CSMA

protocol requires

‘listen’ before

‘transmit’ and

constant ‘listen’

at 10mA current

• Gives lifetime <

10 days with no

transmission

Source: Shnayder, V et al. SenSys 2004

Page 43: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

43

Sensys network architecture

• All links are one-hop

• AP uses one channel; Repeater uses a different channel

APRepeater1000’

100’

Page 44: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

44

Sensys TDMA Protocol (1/2)

125 msec frame

SyncTS0 TS1 TS2 TS3 TS62 TS63

TX

VSN ATime slot 1

VSN B Time slot 3

TX

125 msec frame divided into 64 time slots

Each sensor in the network gets to transmit in one time slot each frame

So latency is less than 125msec

Timeslots permit 50 bytes of data

Page 45: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

45

Sensys TDMA Protocol (2/2)

• Time-slot 0 is the sync timeslot sent from the AP to the

sensors; this timeslot also contains global parameters such

as the detection parameters, the transmit interval, etc.

• Time-slot 2 is used for software download

• Time-slots 4 and 34 are used for ACK packets from the AP

to the sensors

• Remaining time-slots can be used by sensors

125 msec frame

SyncTS0 TS1 TS2 TS3 TS62 TS63

Page 46: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

46

Conclusion

Wireless sensor networks for traffic

measurement are promising and pose

interesting problems of

– Signal processing

– Power consumption

– Engineering design

Page 47: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

47

Outline

• Traffic measurement

• Wireless Sensor Networks

• Vehicle Count, Occupancy, Speed

• Vehicle Classification

• Vehicle Re-identification

• Communication protocol

• Future work

Page 48: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

48

Future work

• Use accelerometers to estimate truck

weight

Page 49: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

49

Equations of motion

• Model pavement as Euler beam with

elastic foundation with a moving load:

Page 50: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

50

A simulation: L=1km, 0/2 = 1.23Hz

• Acceleration peak ~ 0.5-1.0 mg

3.5 Hz

3.7 Hz

-3.1

-13.8

9.8 Hz

4.7 Hz

2.8

14.2

Page 51: A wireless sensor network for traffic surveillance - LAAS · A wireless sensor network for traffic surveillance ... Inductive loop Video Microwave radar ... • Time-slot 2 is used

51

Conclusions

• Most research in sensor networks is

generic: platform, protocols

• Applications require specific designs and

great deal of engineering

• TDMA-based protocols offer great

advantages over random access protocols

• Need to combine sensing with control